Skip to content

PTQ-Level Efficiency, QAT-Level Accuracy | 训练后量化的效率,量化感知训练的精度

Commits on main

20x for awq and 4x for gptq packing speedup (#459)
Wenhua Cheng committed 1 year ago
c803ce9
Support force loading into autoround Format (#453)
WeiweiZhang1 committed 1 year ago
a79d1a3
awq exporting bugfix (#456)
WeiweiZhang1 committed 1 year ago
5ccdb0d
update eval (#450)
Heng Guo committed 1 year ago
d51bd5e
update readme (#455)
Wenhua Cheng committed 1 year ago
6bc0333
fix critic bug of mxfp4 (#451)
Wenhua Cheng committed 1 year ago
abe69c5
step-1 support naive double quant in tuning (#442)
Wenhua Cheng committed 1 year ago
66adfb1
bump to v0.4.6 (#449)
Sun, Xuehao committed 1 year ago
1a9205a
set torch compile to false by default (#447)
WeiweiZhang1 committed 1 year ago
b650162
Refine funcs (#446)
WeiweiZhang1 committed 1 year ago
3a14328
fix bug of evaluate user model (#444)
Heng Guo committed 1 year ago
9aa8803
remove tricky setting in mxfp4 (#445)
Wenhua Cheng committed 1 year ago
1ae5031
19601d4
7d27f67
rm gc collect in packing (#438)
Wenhua Cheng committed 1 year ago
082f01e
fix nblocks issues (#432)
Wenhua Cheng committed 1 year ago
edc2b03
bump to v0.4.5 (#425)
Sun, Xuehao committed 1 year ago
096bc19
fix eval trust_remote_code (#424)
Heng Guo committed 1 year ago
5c075a9
fix cuda UT torch_dtype (#423)
WeiweiZhang1 committed 1 year ago
db8088b
6a49304
22315cc
enable backup pile dataset loading (#417)
WeiweiZhang1 committed 1 year ago
a9b5ae4
support bf16 inference (#420)
Wenhua Cheng committed 1 year ago
29c5119
support naive multi-card tuning (#415)
Wenhua Cheng committed 1 year ago
09f1ab4
937d019
fix log bug and device "auto" bug (#409)
Heng Guo committed 1 year ago
ba5975c
update format readme (#411)
Wenhua Cheng committed 1 year ago
a40ce88
86767b0
support awq with qbits, only support sym (#402)
Wenhua Cheng committed 1 year ago
ba2426c